File:Characterizing ship navigation patterns using Automatic Identification System (AIS) data in the Baltic Sea (IA characterizingsh1094558272).pdf

The Intelligence, Surveillance, and Reconnaissance (ISR) community is interested in developing a model that can assist in characterizing patterns of ship navigation. We examine techniques used to highlight those patterns using historical Automatic Identification System (AIS) data in the Baltic Sea from January to April 2014. A regression model is used to determine which factors influence the amount of time a cargo ship spends in a port in the Saint Petersburg, Russia, area. We find that the best model is able to explain about 29 percent of the variance of the length of time that a vessel is in the Saint Petersburg area. We use three random forest models, that differ in their use of past information, to predict a vessel’s next port of visit. The random forest models we use in this analysis demonstrate that predicting a vessel’s next port of call is not a Markov model but a higher-order network where past information is used to more accurately predict the future state. The transitional probabilities change when predictor variables are added that reach deeper into the past. Our findings suggest that successful prediction of the movement of a vessel depends on having accurate information on its recent history.

Subjects: AIS; Baltic Sea; data analysis; regression; higher-order network; Bayes information criterion; random forest